ReconAgent

ReconAgent is a finance-operations reconciliation backend for invoice, payment, and ledger workflows. It is intentionally scoped to the subniche of AI-native finance automation: AP/AR reconciliation, treasury approvals, exception handling, and auditability.

The portfolio thesis is:

Finance-ops automation companies share a repeated backend/AI bottleneck: reconciling messy financial records, detecting risky exceptions, routing approvals, and preserving audit evidence.

What It Does

1. Imports invoices, payments, ledger entries, and finance policies from CSV.
2. Normalizes vendors, references, dates, currencies, and integer money amounts.
3. Runs deterministic invoice/payment/ledger matching.
4. Detects finance exceptions: duplicate payments, amount mismatches, missing POs, changed bank accounts, overdue invoices, and unmatched ledger entries.
5. Retrieves relevant finance policy text.
6. Produces an explainable proposed action with a deterministic fallback and optional direct LLM integration.
7. Requires human approval for risky actions.
8. Writes hash-chained audit events.
9. Generates eval metrics from labeled reconciliation test cases.

Why This Is Not A Generic AI Project

ReconAgent is built around a repeated finance-ops problem shared by companies such as Round Treasury, Upflow, Moss, Payhawk, Pleo, Spendesk, Mollie, and adjacent lending/document-risk companies such as Crediflow.

The AI is not the source of financial truth. Deterministic matching owns truth; AI explains exceptions and proposes actions inside approval and audit boundaries.

Quickstart

git clone https://github.com/edgarcardenas7/reconagent.git
cd reconagent
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
python -m pytest
uvicorn --app-dir src reconagent.main:app --reload

Open http://127.0.0.1:8000/docs for the API.

Demo Flow

Run these in another terminal while the API server is running:

curl -F "file=@demo/sample-data/invoices.csv" http://127.0.0.1:8000/api/v1/imports/invoices
curl -F "file=@demo/sample-data/payments.csv" http://127.0.0.1:8000/api/v1/imports/payments
curl -F "file=@demo/sample-data/ledger_entries.csv" http://127.0.0.1:8000/api/v1/imports/ledger-entries
curl -F "file=@demo/sample-data/policies.csv" http://127.0.0.1:8000/api/v1/imports/policies
curl -X POST http://127.0.0.1:8000/api/v1/reconciliation-runs
curl http://127.0.0.1:8000/api/v1/exceptions
curl http://127.0.0.1:8000/api/v1/metrics/summary

Technical Docs

• docs/PRODUCT_THESIS.md: portfolio thesis, target company mapping, and system positioning.
• docs/company-research.md: finance-ops company research behind the project scope.
• docs/job-bulletpoint-mapping.md: mapping from common backend/AI role expectations to this repo.
• docs/ARCHITECTURE.md: architecture decisions and service responsibilities.
• docs/FAILURE_MODES.md: operational risks, mitigations, and explicit v1 gaps.
• docs/EVALS.md: implemented metrics and why denominators are reported next to percentages.